Will Tort Reform Bend the Cost Curve? Evidence from Texasjels_1251 173..216

bs_bs_banner Journal of Empirical Legal Studies Volume 9, Issue 2, 173 216, June 2012 Will Tort Reform Bend the Cost Curve? Evidence from Texasjels_1251 173..216 Myungho Paik, Bernard S. Black, David A. Hyman, and Charles Silver* Will tort reform bend the cost curve? Health-care providers and tort reform advocates insist the answer is yes. They claim that defensive medicine is responsible for hundreds of billions of dollars in health-care spending every year. If providers and reform advocates are right, once damages are capped and lawsuits are otherwise restricted, defensive medicine, and thus overall health-care spending, will fall substantially. We study how Medicare spending changed after Texas adopted comprehensive tort reform in 2003, including a strict damages cap. We compare Medicare spending in Texas counties with high claim rates (high risk) to spending in Texas counties with low claim rates (low risk), since tort reform should have a greater impact on physician incentives in high-risk counties. Pre-reform, Medicare spending levels and trends were similar in high- and low-risk counties. Post-reform, we find no evidence that spending levels or trends in high-risk counties declined relative to low-risk counties and some evidence of increased physician spending in high-risk counties. We also compare spending trends in Texas to national trends, and find no evidence of reduced spending in Texas post-reform, and some evidence that physician spending rose in Texas relative to control states. In sum, we find no evidence that Texas s tort reforms bent the cost curve downward. I. Introduction Health-care professionals believe that fear of medical malpractice ( med mal ) litigation motivates widespread overtreatment (defensive medicine). 1 For example, in a recent survey of 1,231 physicians, [a]n overwhelming majority of respondents (91.0%) reported believing that physicians order more tests and procedures than needed to protect themselves *Paik is Research Associate, Northwestern University Law School; email: m-paik@law.northwestern.edu. Black is Chabraja Professor at Northwestern University, Law School and Kellogg School of Management; email: bblack@ northwestern.edu. Hyman is Richard and Marie Corman Professor of Law and Professor of Medicine, University of Illinois, College of Law, 504 E. Pennsylvania Ave., Champaign, IL 61820; email: dhyman@law.uiuc.edu. Silver is McDonald Endowed Chair in Civil Procedure, University of Texas Law School; email: csilver@law.utexas.edu. We thank David Dranove, Craig Garthwaite, Kate Litvak, Anup Malani, Y. Tony Yang, an anonymous referee, and participants in the 2010 ASLME Health Law Professors conference, 2010 Midwest Law and Economics Association, 2010 Canadian Law and Economics Association, 2011 Midwest Health Economics, and 2011 American Law and Economics Association annual meetings, and 2011 Conference on Empirical Legal Studies for helpful comments. We also thank Elliot Fisher for providing the Dartmouth Atlas data on Medicare spending at the HSA level and the Searle Center on Law, Regulation and Economic Growth at Northwestern Law School for financial support. 1 See, for example, Kachalia and Mello (2011). 173

174 Paik et al. from malpractice suits. 2 Physicians also believe that tort reform has great potential to lower health-care spending, by reducing defensive medicine. In the same survey, 90.7 percent of the responding physicians thought that protections against unwarranted malpractice suits are needed to decrease the unnecessary use of diagnostic tests. 3 The true extent of defensive medicine is unclear, but estimates in the $100 $400 billion range (4 15 percent of total health-care spending) are common and higher estimates are sometimes made. 4 Even in Washington, this is real money so calls for tort reform are frequently heard. The House of Representatives passed a cap on non-economic damages ( non-econ cap ) in 2002, 2003, and 2005, only to see them die each time in the Senate. Minority Leader John Boehner included a variety of tort reforms, including a strict cap on non-economic damages, in his 2009 alternative to the Democrats health reform bill. In the Senate, Senator John Cornyn of Texas introduced the Medical Liability Reform Act of 2009, which contained similar provisions. 5 Medical societies lined up in support of these proposals and, for a time, seemed to have convinced the president. In a fall 2009 speech, President Obama stated, I don t believe malpractice reform is a silver bullet, but I have talked to enough doctors to know that defensive medicine may be contributing to unnecessary costs. 6 In the end, the 2010 health-care reform law contained no substantive tort reforms. 7 The House of Representatives again passed a non-econ cap in 2012. 8 Did Congress miss an opportunity to save hundreds of billions of dollars? Can states significantly reduce health-care costs by adopting state-level reforms? Most prior studies of whether malpractice reform affects health-care spending use state-level reform as an (assumed) exogenous shock to malpractice risk, and estimate the impact on spending using state-level data. In contrast, we rely principally on variation in malpractice risk within Texas. Our approach provides a different instrument for assessing whether and how malpractice reform impacts spending. It also provides a larger sample size and less potential for unobserved confounding variables, and thus potentially a stronger basis for causal inference, albeit only for a single (but large and diverse) state. We study Texas because of a fortunate confluence of events. First, Texas adopted tort reform in 2003, which had a dramatic impact on malpractice risk. The Texas reform produced an estimated 60 percent drop in claim rates and a one-third drop in payouts per claim, for a combined drop of over 70 percent in total payouts. Second, Texas is one of only 2 Bishop et al. (2010); see also Studdert et al. (2005). 3 Bishop et al. (2010). Id. 4 Jackson Healthcare (2010) ($650-850 billion/year); Pricewaterhouse Coopers Health Research Institute (undated, approx. 2007) ($210 billion/year). 5 See <http://www.govtrack.us/congress/billtext.xpd?bill=s111-1734>. 6 Obama (2009). 7 Patient Protection and Affordable Care Act 10607 (2010). 8 See H.R. 5, Protecting Access to Healthcare Act <http://www.govtrack.us/congress/bills/112/hr5>.

Will Tort Reform Bend the Cost Curve? 175 two states (Florida is the other one) in which detailed intrastate data on malpractice claims are publicly available. Our core causal inference assumption is that physicians are sensitive to the local risk of a malpractice claim. If defensive medicine is an important spur to spending, then spending within Texas should decline more in counties that had high pre-reform claim rates (high-risk counties), and therefore could expect a larger post-reform drop in claim rates, relative to counties with low pre-reform claim rates (low-risk counties). Medicare uses an administered pricing system, with prices largely set on a national basis. These prices are not, so far as we know, adjusted in response to local tort reform. Thus, we effectively study whether tort reform changed the quantity of medical services provided. Over the long run in a well-functioning market, providers savings from lower malpractice premiums should be reflected in lower health-care prices and thus health insurance rates. We cannot assess with our data whether any such change took place in private insurance markets in Texas. However, for our research question, which is how tort reform affects physician decisions, the quantity of medical services, rather than prices, is what we want to measure. Happily, Medicare spending provides such a measure. We find no evidence that prior to reform, Medicare spending or spending trends were higher in high-risk counties, nor evidence that tort reform reduced Medicare spending or trends in high-risk counties, in each case relative to low-risk counties. Moreover, we can place fairly tight confidence bounds around this null result. In our principal county fixed effects specification (Table 2, regression 2), a one standard deviation increase in malpractice risk predicts a nearly zero ( 0.04%) change in Medicare spending per beneficiary (t = 0.08), with a 95 percent confidence interval (CI) of [ 1.0%, +1.0%]. We also find no significant effect of tort reform on spending for imaging and laboratory services, which is widely considered to be the area of medical practice that is most sensitive to liability risk (predicted change in Table 4, regression 4, of -0.067%; t = 0.85; 95% CI = [ 2.2%, + 0.9%]). These null results are robust to an array of checks with alternative specifications. Turning from spending levels to spending trends, we find no evidence of differing pre-reform trends between high- and low-risk counties, and no evidence that reform reduced spending growth rates in high-risk counties. On the contrary, we find some evidence that Medicare Part B spending (physician spending) trends rose in high-risk counties relative to low-risk counties. In a secondary analysis, we examine whether spending in Texas as a whole changes relative to states that did not undergo tort reform at around the same time. We find no evidence that the 2003 reforms reduced either Medicare spending levels or spending trends in Texas. On the contrary, we find some evidence that physician spending rose in Texas after reform, relative to other states. In sum, no matter how we slice the data, we find no evidence that the Texas 2003 tort reforms bent the cost curve downward, and some evidence of higher post-reform spending by Texas physicians who practice in high-risk counties. Our null result is broadly consistent with other studies of changes in health-care spending after tort reform, most of which rely on state-level data. Some recent state-level studies find no significant effect of tort reform on health-care spending; others find a modest 1 3 percent drop in health-care spending. One early study finds a 4 5 percent drop for heart attack patients. This accumulation of

176 Paik et al. recent evidence finding zero or small effects suggests that it is time for policymakers to abandon the hope that tort reform can be a major element in health-care cost control. Section II reviews the literature on the impact of tort reform on health-care spending and defensive medicine. Section III describes our data sets, hypotheses, and empirical methodology. Section IV studies the impact of Texas 2003 tort reform on medical spending levels. Section V examines the impact of the reforms on spending trends. Section VI discusses our findings. Section VII concludes. II. Literature Review A. What is Defensive Medicine? Tort reform can affect health-care spending both directly and, potentially, indirectly. The direct costs of liability include the cost of malpractice awards, all legal costs associated with the claims, and the administrative costs borne by medical malpractice insurers. 9 Several studies find that tort reforms lead to lower direct costs (approximated by liability insurance premiums). 10 However, because direct costs account for a small fraction (likely under 1 percent) of health-care spending, a drop in direct costs has limited potential to make health-care cheaper. 11 Even a 50 percent drop in direct costs will reduce spending by less than 1 percent and the states that have already adopted tort reform have captured most of these savings. Indirect costs are incurred when providers take steps that do not benefit patients (or lack sufficient benefit to justify their costs) to reduce the likelihood that the providers will be sued. These extra steps are commonly called defensive medicine. Because tort reforms make lawsuits less likely and less expensive, they may reduce defensive medicine and thereby reduce health-care spending. Tort reform advocates and health-care providers often claim that these potential savings are large. Whether tort reform would actually lead to large reductions in indirect costs is unclear. First, as described below, most prior studies find either small effects, or no statistically significant effect on spending from tort reform. These weak results could arise from several sources. First, if costs attributable to defensive medicine are small, then any savings from tort reform must be small as well. Second, providers may practice defensive medicine (assuming they do) for reasons tort reform does not address. For example, physicians may fear second-guessing of their professional judgment whether through lawsuits or peer assessment and practice defensively to avoid such assessments, without much regard to the probability of a negative assessment. 12 Concerns about reputation and 9 Hellinger and Encinosa (2006). 10 Zuckerman et al. (1990); Sloan et al. (1989); Danzon (1984, 1986). 11 Mello et al. (2010) estimate direct costs in 2010 at $10 billion, or 0.4 percent of total health-care spending. 12 Sclar and Housman (2003).

Will Tort Reform Bend the Cost Curve? 177 professional discipline may also motivate defensive practices. Other factors (local practice patterns; the desire to be thorough; economic incentives to run more tests; patient preferences for a definitive answer) can also contribute to unnecessary spending. If most physicians would be inclined to do more for multiple reasons, tort reform could have only a modest impact on spending. Third, tort reform might cause health-care costs to rise. Some defensive medicine takes the form of omission (also called avoidance behavior). Legal risk may cause providers to refrain from performing risky procedures. Tort reform could encourage greater use of such procedures. Montanera develops a model in which physicians do more for some patients, but avoid others, leading to a nonlinear effect of malpractice pressure on healthcare spending. 13 When asked in surveys, many physicians report limiting risky procedures due to fear of malpractice liability. 14 Similarly, if physicians have economic incentives to provide services, legal risk could provide a counterweight against this physician-induced demand. Tort reform weakens this counterweight. These higher spending channels could offset any savings from performing fewer defensive procedures in response to med mal risk. 15 This freeing effect could explain our finding some evidence that physician spending rose after reform in high-risk counties. Finally, some studies find evidence that malpractice liability leads to improved quality. 16 Tort reform could cause care quality to decline, which could, in turn, cause spending to rise. Thus, the overall impact of tort reform on spending cannot be answered on theoretical grounds. Both the sign and magnitude of any effect must be determined through empirical study. B. How Much Defensive Medicine is There, and How Much Does it Cost? 1. Studies Exploiting Plausibly Exogenous Variation in Risk Kessler and McClellan performed the first rigorous studies of the impact of tort reforms on health-care spending. 17 Using longitudinal data on Medicare beneficiaries who received cardiac treatments in hospitals for acute myocardial infarction (heart attack) or ischemic heart disease in three years (1984, 1987, and 1990), they found that damages caps and other reforms that directly limited liability reduced posttreatment medical spending by 5 9 13 Montanera (2011). 14 In a 2010 survey of Illinois physicians, 89 percent reported that malpractice fears caused them to order more tests than medically needed, but 66 percent also reported that they reduced or eliminated high-risk services or procedures, and 77 percent either did so or planned to. ISMIE Mutual Insurance Company and Illinois State Medical Society (2010). 15 Currie and MacLeod (2008). 16 See, e.g., Dhankhar et al. (2007); Lakdawalla and Seabury (2009); Zabinski and Black (2012). 17 Kessler and McClellan (1996).

178 Paik et al. percent, without adverse health effects. Other tort reforms did not produce statistically significant spending reductions. In response to criticisms that their study had not controlled for managed care penetration, Kessler and McClellan reanalyzed their data with this control and found a 4 5 percent decline. 18 In their original article, Kessler and McClellan observed that if our results are generalizable to other medical expenditures outside the hospital, to other illnesses, and to younger patients, then direct [tort] reforms could lead to expenditure reductions of well over $50 billion per year without serious adverse consequences for health outcomes (emphasis added). Tort reform advocates played up Kessler and McClellan s number and played down their qualifications. One stated flatly that nationwide tort reform would save well over $50 billion a year. 19 In 2002, the Department of Health and Human Services issued a report that relied on Kessler and McClellan s first article and concluded that tort reform would save $60 $108 billion in health care costs each year. 20 President Bush gave several speeches that relied on this figure. More recent studies cast doubt on the generalizability of Kessler and McClellan s results. The Congressional Budget Office applied their methods to a broader range of medical conditions, and found no evidence that restrictions on tort liability reduce medical spending. 21 A follow-up CBO study finds that states that implemented tort reforms in the 1980s had above-average health-care pricing before the 1983 implementation of the Medicare Prospective Payment System, which disproportionately affected states with higher pricing. When CBO controls for this (which Kessler and McClellan did not), its estimate of the effect of a cap on non-economic damages on Medicare spending is a statistically insignificant-1.6 percent. 22 A study by Sloan and Shadle that covers more conditions and more years than Kessler and McClellan also found insignificant results. 23 Lakdawalla and Seabury used county-level non-economic damage awards as an instrument for med mal risk. They found that lower risk predicts modestly lower health-care prices, no significant change in health-care quantity, and somewhat higher mortality. 24 18 Kessler and McClellan (2002). 19 Huber (1997). 20 U.S. Department of Health and Human Services (2002). 21 CBO (2004). 22 CBO (2006). In 2009, the CBO estimated, based on a survey that included more recent studies, that a package of tort reforms, including caps on non-economic and punitive damages, would reduce health-care spending by 0.5 percent, comprised of a 0.2 percent decline due to lower direct med mal costs and a 0.3 percent decline from slightly less utilization of health care services. CBO (2009). 23 Sloan and Shadle (2009). 24 Lakdawalla and Seabury (2009) (10 percent increase in malpractice awards predicts 1.2 percent higher spending, due to higher prices, no change in quantity, and 0.2 percent lower mortality). The Lakdawalla and Seabury identification strategy is similar to ours in relying on county-level rather than state-level measures, and thus assuming that physicians know about and are sensitive to local risk. Their strategy assumes either (implausibly, in our view) that

Will Tort Reform Bend the Cost Curve? 179 Studies of childbirth treatment and outcomes have also produced mixed results. Currie and MacLeod found that non-econ caps increased cesarean section rates (thus increasing costs), while joint and several liability reform decreased cesarean section rates (thereby reducing costs). 25 Frakes examined the impact of non-econ caps on cesarean rates, episiotomies, and bed days post-delivery. 26 Non-econ caps predict roughly 5 percent fewer episiotomies (depending on specification) and perhaps also a 3 percent or so decline in hospital stays, but did not affect cesarean rates. In contrast, Yang et al. find that non-econ caps predict lower primary and total cesarean rates, and higher VBAC (vaginal birth after cesarean) rates. 27 Finally, recent studies by Dranove and Watanabe and by Gimm assess whether Florida obstetricians practice patterns changed after they were sued. 28 Dranove and Watanabe found lagged increases in cesarean section rates following hospitals and physicians involvement in claims, but the effects were small and temporary. Gimm finds insignificant effects. Overall, these studies suggest that tort reforms have limited potential to contain health-care costs by reducing defensive medicine among obstetricians. Avraham, Dafny, and Schanzenbach measured spending in terms of premiums collected by employer-funded health insurance plans representing over 10 million Americans annually from 1998 and 2006. 29 They found that each of three reforms (non-econ caps, abrogation of the collateral source rule, and reform of joint and several liability) reduced premiums for self-funded health plans by 1 2 percent, but had no effect on premiums for fully insured plans. 2. Studies Without Exogenous Variation in Risk The studies discussed below lack exogenous variation in risk, so provide a weaker basis for causal inference. We review them briefly for completeness. Hellinger and Encinosa rely on state tort reform as a measure of med mal risk, but have only cross-sectional data. They estimate that a state cap on non-economic damages predicts 3 4 percent lower Medicare spending. 30 Morrissey, Kilgore, and Nelson have panel physicians are informed about changes in county-level non-economic damage awards or (more plausibly) that (1) higher awards lead to more suits, and (2) physicians are aware of the local risk of being sued. In contrast, we directly measure the impact of the Texas reforms on claim and payout rates; thus, we do not need to assume an indirect effect on claim rates through the channel of non-economic damage awards. 25 Currie and McLeod (2008). 26 Frakes (2010a, 2010b). 27 Yang et al. (2009). 28 Dranove and Watanabe (2010); Gimm (2010). An earlier study of Florida obstetricians by Grant and McInnes (2004) found a 1 percent increase in cesarean rates for obstetricians who experienced a large payout claim during 1992 1995, but no significant change for obstetricians who experienced smaller claims. 29 Avraham et al. (2012). 30 Hellinger and Encinosa (2006).

180 Paik et al. data, and find no effect of damage caps on private employer premia using a state fixed effects specification, but their time period is short (1999 2004) and effectively predates most of the caps adopted in the 2000s. 31 Another plausible measure of med mal risk is med mal insurance premia. In two studies, Baicker and co-authors find no overall association between insurance premia and Medicare spending, but do find an association for the Medicare Part B spending subcategory for diagnostic, laboratory, and x-ray services. 32 They hypothesize that diagnostic testing is more likely to reflect defensive medicine than medical practice more generally, and thus is more sensitive to liability risk. Thomas, Ziller, and Thayer compare medical spending for 35 clinical specialties in regions with high and low med mal insurance premiums. They conclude that even a large (30 percent) reduction in malpractice premiums would predict only a 0.4 percent decline in health-care spending. 33 For childbirth, Dubay, Kaestner, and Waidmann use county fixed effects and find a statistically significant but small association between obstetrician malpractice premiums and cesarean section rates. 34 Localio and co-authors found an association between malpractice risk and cesarean rates in New York State. 35 However, Baldwin and co-authors found no relationship between cesarean rates and either personal malpractice experience or countylevel malpractice risk within Washington State. 36 Another potential risk measure is rates for physicians paid claims, reported to the National Practitioner Data Bank. Dhankhar, Khan, and Bagga find that higher claim rates are associated with improved post-heart-attack outcomes, and predict lower treatment costs for less severely ill patients. 37 C. Physician Perception and Malpractice Risk The logic of tort reform is straightforward: if physicians respond to liability risk by practicing defensive medicine, tort reform that lowers liability risk will also reduce defensive medicine. It is the perception of liability risk (and not actual risk) that influences physicians propensity to practice defensive medicine. Thus, tort reform may not reduce defensive medicine if it does not change physicians perception of the liability climate even if reform does, in fact, reduce liability risk. Several studies indicate that physicians dramatically overestimate their liability risk. For example, a study of physicians in New York in 1984 31 Morrissey et al. (2008). 32 Baicker and Chandra (2005); Baicker et al. (2007). 33 Thomas et al. (2010). The authors do not report the statistical significance of their results. 34 Dubay et al. (1999). 35 Localio et al. (1993). 36 Baldwin et al. (1995). 37 Dhankhar et al. (2007). This study includes reasonably detailed covariates for hospitals and patients but, oddly, none for states.

Will Tort Reform Bend the Cost Curve? 181 indicated that they overestimated their risk of being sued by a factor of three. 38 Physician risk perceptions do vary with the level of med mal risk, but not nearly as much as they likely should. 39 III. Data, Hypotheses, and Methodology A. Intrastate Causal Inference Strategy Most prior studies that estimate the causal effect of tort reform on health-care spending or outcomes do so using state med mal reforms as an exogenous shock to med mal risk, with state-level spending as the dependent variable. One problem with this approach is that state reforms differ substantially. For example, damage caps vary in what type of damages they apply to (total vs. non-economic vs. punitive), the cap level, whether the cap varies with the type and number of defendants, and whether the cap level is indexed for inflation. Even if one studies only one type of damages cap, there are obvious difficulties with treating all caps as identical, as all previous studies do. Second, reforms are often bundled together, which complicates any attempt to estimate the impact of one particular type of reform, such as a non-econ cap. Third, there are only so many state-level tort reforms, so sample sizes are small. The principal methodological innovation in this study is to rely on a large tort reform shock to a single state (Texas), posit (and verify) that this reform reduced local med mal risk more strongly in areas that had higher pre-reform risk levels, and look for intrastate differences in how that shock to local risk affected local spending. This approach is attractive because Texas experienced a uniquely large med mal shock, it has detailed county-level data on med mal risk, and that risk varies substantially across counties. We can study roughly 200 Texas counties, all subject to the same reform, instead of a much smaller number of states, each subject to a different package of reforms. To estimate the impact of Texas tort reform on health-care spending, we need to make two core causal inference (identification) assumptions, which we cannot directly test. First, we assume that providers are sensitive to local med mal risk, and providers in high-risk counties will perceive a larger reduction in med mal risk than providers in low-risk counties. One channel for this differential impact is the absolute level of malpractice insurance premiums. Suppose that tort reform reduces the likelihood of a med mal claim by 50 percent from its pre-reform level. This might imply a $50,000 drop in annual premiums in a high-risk county, say from $100,000 to $50,000, but only a $15,000 drop in a low-risk county, from $30,000 to $15,000. A second channel is conversations between doctors about personal experiences in being sued. Here, too, a similar percentage drop in claim frequency should affect personal experience more in high-risk than low-risk counties. 38 Lawthers et al. (1992). 39 Carrier et al. (2010).

182 Paik et al. Second, we assume that any other statewide factors that influence health-care spending affect high-risk and low-risk counties similarly. The low-risk counties thus serve as a control group for the high-risk counties, similar to a difference-in-differences (DiD) research design. 40 We also assume that any impact of med mal reform on health-care spending will appear within a reasonable period of time. The longer the lag, of course, the less confidence one can have that there are not other, unobserved differences that emerge over time between high-risk and low-risk counties, which would affect the validity of our DiD-like research design. We study lags of up to four years. We asked physicians, including health outcomes researchers, and the proponents of the Texas reforms whether these assumptions seemed reasonable. They generally concurred that physicians are likely to be aware of and respond to local risk. Their principal concern was a lag between reform and response, due to the stickiness of local practice patterns. 41 B. Interstate Analysis We also assess whether tort reform predicts any change in Texas spending levels, or spending trends, relative to other states. We use three principal control groups: (1) the 41 other states that did not undergo major tort reform during our principal 1999 2009 sample period; (2) the 19 other states that do not have caps on non-economic or total damages at all (no-cap states); and (3) a subset of nine no-cap states that are geographically and culturally relatively similar to Texas. The core causal inference assumption for this analysis is that tort reform is the only factor that causes Texas spending to change, relative to the control states. Because this assumption is untestable and could easily be wrong, we consider this to be a secondary analysis, akin to a robustness check on our primary intrastate analysis. If Texas spending had fallen relative to control states, this would suggest a possible statewide impact of reform not captured by our analysis of intrastate variation in risk. Conversely, if Texas spending remains roughly constant or rises relative to control states, this is consistent with our main intrastate finding that a larger med mal risk shock does not predict lower post-reform spending. C. Data Sources and Variables 1. Medicare Spending Data and Variables Our Medicare spending data cover 1999 2009, and come from the Centers for Medicare and Medicaid Services. 42 This source provides annual spending at the county level; there are 40 Compare Atanasov et al. (2010), who assess the impact of Bulgarian legal reforms that limit self-dealing on firm market values by comparing firms with high self-dealing risk to low-risk firms. 41 For example, one of us (Black) presented the research design to the Colorado Cardiovascular Outcomes Research Group <http://www.coloradooutcomes.org/> (May 2011). 42 Currently, data from 2001 to 2009 are available at <http://www.cms.gov/medicareadvtgspecratestats/05_ffs_ Data.asp>. We have data from this source from 1999 (downloaded before it rolled off). We thank Leemore Dafny and

Will Tort Reform Bend the Cost Curve? 183 254 counties in Texas. It reports total Medicare spending, Part A spending, and Part B spending, but not spending within subcategories of Part B. We therefore use another data set from the Dartmouth Atlas of Health Care, which provides annual Medicare spending through 2007, including subcategories of Part B, at the Hospital Service Area (HSA) level; there are 205 HSAs in Texas. An HSA is a local health-care market, usually centered around a single hospital. 43 We use the fraction of overall HSA-level Part B spending in each Part B subcategory to estimate spending in each Part B subcategory by county. We map HSAs onto counties using a zip-code-county-hsa cross-walk and population weights. Appendix A provides details on our estimation procedure. Our principal dependent variable is ln(relative Medicare spending), defined as ln(texas spending per Medicare enrollee/mean state spending per enrollee in control states). The principal control group is 41 states that did not adopt a cap on non-economic or total damages (a damages cap) during our sample period. 44 In robustness checks we also use a non-reformed control group of 19 states that have never adopted a damages cap, and a narrow control group of nine non-reformed states that excludes northeastern and eastern states, leaving states that are likely to be culturally closer to Texas. 45 2. Medical Malpractice Data Our data on med mal claims come from the Texas Closed Claims Database (TCCD), a publicly accessible database maintained by the TDI. 46 From the larger database we extracted claims that closed from 1990 2009 with payouts of at least $10,000 involving medical Seth Seabury for sharing their data for earlier years with us. We use Medicare data back to 1988 (when our malpractice data begin) in unreported robustness checks, with results consistent with those we report. We report results using only data from 1999 forward due to large, implausible changes in Texas Medicare spending in some earlier years (e.g., a 70 percent jump in Part B spending per enrollee in 1996 vs. 1995, and then an 18 percent drop in 1998). 43 Data from 1992 to 2007 are available at <http://www.dartmouthatlas.org/data/download.shtm>. After 2007, Dartmouth Atlas provides a new data series, which does not include the Medicare Part B subcategories we study here. 44 The 41 states without recent reforms are 50 states plus District of Columbia, less Texas and eight other states that adopted damage caps over 2003 2006 (Florida, Georgia, Illinois, Mississippi, Nevada, Ohio, Oklahoma, and South Carolina), and Oregon, in which a cap was invalidated in 2000. No other states adopted or repealed caps on total or non-economic damages during our sample period. Alaska lowered its non-econ cap from $400k to $250k in 2006; Wisconsin raised its non-econ cap from $450k to $750k in 2006. Our definition of a damages cap excludes caps on punitive damages, on the grounds that these caps likely have only a small impact on med mal risk. 45 The 19 non-reformed states are Alabama, Arizona, Arkansas, Connecticut, Delaware, District of Columbia, Iowa, Kentucky, Maine, Minnesota, New Hampshire, New York, North Carolina, Pennsylvania, Rhode Island, Tennessee, Vermont, Washington, and Wyoming. The nine similar non-reform states are Alabama, Arizona, Arkansas, Iowa, Kentucky, Minnesota, Tennessee, Washington, and Wyoming. 46 This database contains individual reports of all personal injury claims closed from 1988 on, covered by five lines of commercial insurance monoline general liability, auto, multiperil, medical professional liability, and other professional liability involving payouts by all defendants of more than $10,000 in nominal dollars. Data are currently available through 2009. TDI checks the reports for internal consistency and reconciles them against aggregate annual reports filed by each insurer. For a fuller discussion of the TCCD, the med mal data set, and data set limitations, see Black et al. (2008). The Texas Department of Insurance (TDI) Closed Claim Reporting Guide (2004) (containing

184 Paik et al. malpractice by physicians or hospitals. 47 Our sample includes 18,981 nonduplicate cases involving total payouts over 1988 2008 of $4.9 billion. The TCCD only includes insured claims. Most physicians carry malpractice insurance, but we lack data on claims against physicians employed by the University of Texas hospital system, which is self-insured, and on self-insured hospitals. We lack data on cases with zero or small payout. 3. Med Mal Risk Variables Our principal med mal risk measure is long-term med-mal risk, defined for each county as the mean over 1999 2003 of ln(1 + (no. of claims/100,000 population)), normalized to mean = 0 and standard deviation = 1. We take a five-year average to capture the notion that physicians are likely to have a general sense for the level of risk they face, but are not likely to change their risk perceptions because of short-term fluctuations in claim rates. We add 1 before taking logs to avoid having to drop counties with zero claims. Before normalization, this measure has mean = 0.86 and standard deviation = 0.71, so there is reasonable variation between counties. We use a number of alternate measures in robustness checks (see Table 5). 4. Control Variables We obtain Texas population by county and year from the U.S. Census Bureau. 48 When mapping counties onto HSAs or vice-versa, we use population weights based on exact Census 2000 zip-code-level data. We define a county as urban or rural based on the USDA definition. 49 5. Texas Med Mal Reform In 2003, Texas capped non-economic damages in med mal cases against physicians and other individual health-care providers at $250,000 nominal ($161,000 in the 1988 dollars we use in this article), with an additional $250,000 nominal possible if a hospital or other reporting instructions), the long and short forms, summary Closed Claim Annual Reports, and the data on which we rely are available at <http://www.tdi.state.tx.us>. 47 The claims in our med mal data set meet at least two of the following criteria: (1) payment under medical professional liability insurance; (2) physician or hospital defendant; (3) injuries caused by complications or misadventures of medical or surgical care. Claims with payout of $10,000 $25,000 are reported on a Short Form ; claims with payout of at least $25,000 are reported on a Long Form. Only the Long Form contains the nature of the injury. We require that Short Form claims meet criteria (1) and (2). The reporting thresholds are not inflation adjusted. Thus, some claims that were reported in later years would not have been reported in earlier years if the thresholds had been inflation adjusted. The larger claims we study account for over 99 percent of total payout on all paid claims. We convert payouts to 1988 dollars using the Consumer Price Index for All Urban Consumers (CPI). Source: <www.bls.gov/cpi/>. 48 The Census provides estimated county population by year, based on interpolation between decennial censuses. The zip-code-level data are available at <http://www.census.gov/epcd/www/zipstats.html#aff>. 49 U.S. Department of Agriculture, 2004 County Typology Codes, available at <http://www.ers.usda.gov/data/ TypologyCodes/>.

health-care institution is also liable, up to a maximum of two institutions, for a maximum overall cap of $750,000 nominal. Other components of the 2003 reforms include making the death cap apply per claim, rather than per defendant, higher proof standards for emergency care, and a requirement that plaintiffs file an expert report within 120 days of suit with regard to each defendant s negligence (by a practicing physician, if the defendant is a physician). 50 As Figures 1 and 3 reflect, these reforms dramatically reduced both claim rates and payouts per claim; this reduction in claim rate was larger in counties with higher pre-reform med mal risk. Thus, these reforms provide a large exogenous shock to the tort system, which we can use to assess whether this large decrease in malpractice risk affects health-care spending. D. Hypotheses Will Tort Reform Bend the Cost Curve? 185 Our hypotheses, based on the defensive medicine literature and on popular views about how physicians respond to med mal risk, are as follows. Hypothesis 1: Med mal risk and local spending. If med mal risk significantly affects health-care spending, then (subject to endogeneity concerns) there should be a positive association between county-level med mal risk and health-care spending. Hypothesis 2: Med mal reform and statewide spending trends. If med mal risk significantly affects health-care spending, then the 2003 tort reform will result in lower health-care spending in Texas, compared to other states. Hypothesis 3: Med mal reform and local spending. Med mal reform will have a greater impact on spending in Texas counties with high pre-reform med mal risk than in low-risk counties because the reforms will cause a larger drop in risk in high-risk counties. Hypothesis 4: Med mal reform and spending on imaging and laboratory services. Med mal reform will have a greater impact on services that are more discretionary (especially laboratory tests and imaging studies) than on services that are less discretionary (hospitalizations). 51 To preview our conclusions, we find no support for any of these hypotheses. Instead, we consistently find null results, with reasonably tight confidence bounds. Where we do find statistically significant results, they have the opposite sign from that predicted. In particular, we find non-robust evidence that Texas physician spending rises after reform (contra to 50 Damages in death cases were limited to $975,000 total; this limit previously applied separately to each defendant. The standard of liability for emergency medical care in a hospital emergency room or obstetrical unit (or in surgery immediately after admission to the emergency room) was raised from ordinary negligence to willful and wanton negligence. The 2003 reforms included several less significant changes, including the following. In general, defendants who are less than 50 percent at fault are liable for damages multiplied by their percentage of fault, with damages reduced by any prior settlements; med mal defendants may choose to have damages reduced by either the amount of prior settlements or the percentage of fault assigned to the settling defendants. Defendants can claim that part or most of the fault belongs to a nondefendant third party, and the jury must allocate fault to that person. And there is a 10-year statute of repose, so if parents do not sue for a child s birth injuries by age 10, the child cannot do so when he or she becomes an adult. 51 See Baicker et al. (2007), discussed in the literature review.

186 Paik et al. Hypothesis 3), with higher spending coming from high-risk counties (again contra to Hypothesis 3). We also find non-robust evidence that imaging and lab spending rises in high-risk counties after reform (contra to Hypothesis 4). E. Empirical Methodology We test these hypotheses using a variety of regression specifications. Our preferred specification for our intrastate hypotheses 3 and 4 uses county fixed effects to control for unobserved local factors that might predict both med mal risk and Medicare spending: ln( relative spending ) = α + C + T + β PR + γ Risk + λ ( PR Risk ) + w controls it + ε it. (1) it i t t i t i Here, 1 indexes counties, t is time, C i and T t are county and (year-and-after) fixed effects, PR t is a post-reform dummy (= 1 in 2004 or after; 0 otherwise), and Risk i is med mal risk. Other controls include ln(population) and urban dummy. 52 The coefficient of principal interest is on the interaction term. In robustness checks, we obtain similar results if we include county-specific time trends, and also with county random effects and pooled OLS models. We address the potential correlation among observations of the same county across years by using standard errors clustered on county (in HSA-level analyses, we cluster on HSA). In robustness checks, we obtain similar, indeed somewhat smaller, standard errors for our OLS results with two-way clustering on year and county. We cannot combine two-way clustering with fixed or random effects. Medicare spending is spending per enrollee. 53 We also map our med mal risk measures onto HSAs and study spending at the HSA level (with HSA fixed effects). The independent variables of principal interest are: various measures of med mal risk, a med mal reform dummy (= 1 in Texas in 2004 and after; 0 for other states and for earlier years), and an interaction between med mal risk and reform dummy. Other controls include ln(population), and either a year trend variable or year dummy variables. Some important limitations of our study: we study only one state, albeit a large and diverse one, which experienced a very large reform shock. We have detailed data only on claims with payout over $10,000 (nominal). For other claims, including those that close without payout, we have only statewide aggregate counts. There is puzzling year-to-year variation in Medicare spending data in the mid-1990s, which leads us to start our pre-reform period only in 1999. Our HSA-level data (and thus our data on subcategories of physician spending) extends only through 2007. 52 Post-reform dummy, our risk measure variable, and urban dummy drop out with county and time fixed effects. We include them in Equation (1) because they are included in some of the alternate regression specifications we use below (e.g., random instead of fixed effects, time-varying risk). 53 In robustness checks, we obtain similar results with ln (Medicare spending per enrollee) as the dependent variable instead of ln(relative spending), and U.S. spending either included as a control variable or omitted, and with nonlogged relative Medicare spending per enrollee as the dependent variable.

Will Tort Reform Bend the Cost Curve? 187 Figure 1: Pre-reform versus post-reform med mal risk. post-reform risk (2004-2008) 0 1 2 3 45 degree 0 1 2 3 pre-reform risk (1999-2003) Note: Scatter plots of 45 degree line and regression line from regression of non-normalized post-reform long-term med mal risk (2004 2008) on non-normalized pre-reform long-term med mal risk (1999 2003) and constant. Regression coefficient = 0.53 (t = 9.67 for null hypothesis that coefficient = 1). F. Exploring the Causal Inference Assumptions 1. Does Pre-reform Risk Predict Post-reform Drop in Risk? Our empirical strategy assumes that (1) high-med-mal-risk regions will experience a larger drop in med mal risk after tort reform than low-risk regions; and (2) physicians are aware of local med mal risk. The first assumption is testable. In Figure 1, we plot non-normalized pre-reform long-term med mal risk (over 1999 2003) on the x-axis, and post-reform long-term med mal risk (over 2004 2008) on the y-axis, together with a regression line showing the relationship between the two, and a 45-degree line. If med mal risk declines more in high-risk counties, we expect the regression line to have a slope less than 1; the actual slope is substantially less than 1 (slope = 0.53; t = 9.67). 54 2. Geographic Variation in Med Mal Risk One might also question our causal inference strategy if particular county (or HSA) factors (e.g., population, urban vs. rural, percent Hispanic, or a particular region within Texas) were dominant predictors of med mal risk. Our preferred fixed effects specification controls for time-invariant factors, but such a factor could still be associated with changes in 54 A partial cause for the slope of less than 1 in Figure 1 is regression to the mean (or median). However, we would not expect doctors to be sophisticated enough to adjust their behavior taking into account the potential for an already long-term risk measure to regress to the mean. Also, the crossing point between the two lines is well below the median (mean) pre-reform risk of 0.79 (0.86), so regression to the mean (or median) is only a partial explanation for the observed results.

188 Paik et al. Figure 2: Geographic distribution of long-term med mal risk. Note: Figure shows county-level long-term med mal risk, defined as (mean over 1999 2003 of ln(1 + (no. of claims)/100,000 population) t). Darker shading indicates higher risk. Medicare spending. 55 Fortunately, this is not the case. Figure 2 shows the geographic distribution of long-term med mal risk by county. Darker shading shows higher risk. The high-risk counties are well distributed all over the map, as it were. In regression analyses, larger population predicts higher risk, but there is substantial scatter that is not explained by population or other demographic controls. 56 IV. Med Mal Reform and Spending Levels A. Aggregate Trends By any measure, Texas 2003 tort reforms transformed the med mal liability environment. As Figure 3 reflects, from 1990 to 2003, claim frequency (the number of paid claims per 55 To give a concrete example, suppose that high-med-mal-risk counties were concentrated in Southeast Texas, near Houston. One would worry that some unobserved change, affecting the Houston area around the time of med mal reform, caused spending to rise, and offset a decrease caused by med mal reform. 56 For example, in the smallest fifth of counties by population, long-term med mal risk has mean = 0.42, s=0.67, while the largest fifth have mean risk = 1.60, s = 0.50.

Will Tort Reform Bend the Cost Curve? 189 Figure 3: Medical malpractice claim rates and payouts in Texas. number of claims 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 Tort reform 30 27 24 21 18 15 12 9 6 3 0 real 2009$ number of claims per 100,000 population med mal payout per capita Note: Number of claims per 100,000 population by year for all claimants (left scale), and payouts per capita (right scale), for 14,995 nonduplicate, non nursing home, med mal cases closed from 1990 2009 with payout >$25,000 in 1988 dollars. Texas tort reform in 2003 is depicted by vertical line. Amounts in 2008 dollars. 100,000 Texas residents) and claim severity (defined as payout per capita for all paid claims that closed in a given year) were generally stable. Post-reform, both frequency and severity trended sharply downward. Over 2003 2009, paid claims/100,000 population fell by 57 percent and payout per large paid claim (over $25,000 in 1988 dollars) dropped by 35 percent. 57 Because the TCCD contains only paid claims and claims typically close several years after being brought, Figure 3 tells only part of the story. One would also like to assess the impact of Texas 2003 tort reforms on the rate at which new claims were filed and on the value of claims currently in the pipeline. Various sources, including the University of Texas Medical System, indicate that new filings declined sharply. 58 In 2009, claims intake at the Texas Medical Liability Trust (TMLT), the state s largest malpractice carrier, was approximately half of the amount [TMLT] experienced in the years prior to tort reform, even though the number of insured physicians increased. Open claims also appear to be worth less than similar claims pending in prior years. In 2009, TMLT s [t]otal trial losses... amounted to less than $1 million, one of the lowest trial loss years ever recorded. 57 Paik et al. (2012). On the post-reform drop in claim frequency and payouts, see also Carter (2006) and Daniels and Martin (2009). On declining med mal premia, see also Guardado (2009) and Slavin (2010). The drop in all claims (including claims with zero or small payout) is similar to the drop in large paid claims shown in Figure 3. 58 Stewart et al. (2011) report an 80 percent drop in lawsuits following surgery, from 40 suits/100,000 procedures before reform to 8 suits/100,000 procedures after reform.

190 Paik et al. Insurance premiums have dropped as well. TMLT reports that the 2003 reforms dropped the cost of medical liability insurance by 50% for its policyholders. 59 One should also note the scale of the right-hand y-axis in Figure 3. By 2007, total med mal payouts were only $7/capita a tiny amount. Even if defense costs and insurer overhead add 50 percent to this number, direct med mal costs are only $10/capita. Manifestly, direct med mal spending is a small contributor to health-care costs. Given the large post-reform drop in med mal risk, Hypothesis 2 predicts a drop in statewide health-care spending. However, we find no evidence of a drop. Figure 4 presents the Texas spending gap (Texas Medicare spending per enrollee - Medicare spending per enrollee in control states, separately for Part A (loosely, hospital spending), Part B (loosely, physician spending), and total Medicare spending (A + B) from 1999 2009. In Panel A, we compare Texas with the 41 U.S. states that did not adopt or repeal damages caps during our sample period. In Panel B, we restrict the control group to the 19 non-reformed states. The Texas spending trend relative to the controls is similar in both panels, and is also similar (figure not shown) if we use the narrow control group of nine similar non-reformed states. The Texas Part A spending gap is small and declines somewhat in 2008 and 2009. The Texas Part B spending gap rises over 2002 2007, with no downward bend after reform. This is not the pattern predicted by Hypothesis 1. This is, to be sure, a weak test. Other factors affect statewide spending trends indeed, the Texas spending gap is not flat prior to reform. Still, these aggregate data suggest that tort reform likely did not have a large impact on health-care spending. We return in Section V to the question of whether Texas spending levels or trends rose or fell after reform. B. Overall Association Between Medicare Spending and Med Mal Risk We turn next to Hypothesis 1 medical spending will be higher in counties with higher med mal risk. We begin with graphical analysis. Both per-capita health-care spending and med mal risk are higher in urban areas. Both effects could be related to population, rather than a causal effect of med mal risk on spending or vice-versa. We therefore regress ln (relative spending) on ln (population), urban dummy, year dummies, and constant, and compute residuals from this regression for each county and year. We similarly regress annual med mal risk, defined as normalized [ln (1 + number of claims/100,000 population)], on the same control variables, and compute residuals. In Figure 5, Panel A, we plot the spending residuals against the risk residuals, and show a regression line from regressing residual spending on residual risk plus constant term. Panel B is similar, except we drop county-year observations with zero med mal claims. In both figures, the slope of the regression line is effectively zero and is not close to being statistically significant. We obtain similar results if we weight observations by ln(population) 59 Texas Medical Liability Trust, 2009 Annual Report, p. 4. This is in nominal dollars; the decline would be larger if adjusted for inflation.

Will Tort Reform Bend the Cost Curve? 191 Figure 4: Texas spending gap, 1999 2009. Panel A. Texas vs. 41 states without recent reforms 1,200 Spending gap (real 2007 $) 1,000 800 600 400 200 0-200 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Part A Part B Total Panel B. Texas vs. 19 non-reformed states 1,200 Spending gap (real 2007$) 1,000 800 600 400 200 0-200 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Part A Part B Total Note: Graphs of Texas spending gap (Texas - control state spending per enrollee) for Medicare Part A, Part B, and total Medicare spending. Medicare Part A covers inpatient care in hospitals and hospice care services. Medicare Part B covers physician services and outpatient care, including home health, imaging, and clinical laboratory testing. Control groups are specified in the text. Texas tort reform in 2003 is depicted by vertical line. Amounts in 2007 dollars.

192 Paik et al. Figure 5: Association between med mal risk and Medicare spending..5 residual -.5 0 A. Including 0 claim counties -2 0 2 4 6 normalized annual med mal risk residual -.5 0.5 B. Excluding 0 claim counties 0 2 4 6 normalized annual med mal risk C. Including 0 claim counties D. Excluding 0 claim counties spending growth rate (1999-2003) -.5 0.5-1 0 1 2 3 long-term med mal risk (1999-2003) spending growth rate (1999-2003) -.5 0.5-1 0 1 2 3 long-term med mal risk (1999-2003) Panels A and B: Annual Med Mal Risk and Spending Levels Panel A. Scatter plot of residuals from a regression of ln(medicare spending per enrollee/u.s. average spending per enrollee) ( relative spending ) on population, urban and year dummies over 1999 2008, against annual med mal risk, defined as normalized [ln(1 + (number of claims/100,000 population)), and regression line for regression of first-stage residuals on normalized annual med mal risk and constant term, for 254 Texas counties over 1999 2008 (n = 2,540). Slope = 0.003 (t = 0.78). Amounts in 2007$. Panel B. Similar scatter plot and regression, after dropping county-year observations with 0 claims (n = 1,040 for 215 counties). Slope = +0.005 (t = 0.50). Panels C and D: Long-Term Med Mal Risk and Spending Trends Panel C. Scatter plot of relative Medicare spending growth rate over 1999 2003, calculated as ln(relative spending) 2003 ln(relative spending) 1999, and long-term med mal risk, defined as normalized [mean over 1999 2003 of ln(1 + (no. of claims)/100,000 population)], and regression line for regression of relative Medicare spending growth rate on long-term med mal risk and constant term, for 254 Texas counties. Slope is -0.006 (t = 0.75). Amounts in 2007$. Panel D. Similar scatter plot and regression line, after dropping 65 counties with 0 claims (n = 189). Slope = 0.008 (t = 0.99). (not reported). There is thus no evidence that higher med mal claim rates are associated with higher spending levels. In Panels C and D of Figure 5 we assess spending trends. Panels C and D plot relative (Texas minus U.S.) Medicare spending growth for each county over 1999 2003 against long-term med mal risk, and show a regression line from regressing spending growth on long-term med mal risk plus constant term. There is no significant association between med mal risk and pre-reform spending trends, whether we include zero-claim counties (Panel C)